Next Article in Journal
A Selection Process for Genetic Algorithm Using Clustering Analysis
Previous Article in Journal
Scheduling Non-Preemptible Jobs to Minimize Peak Demand
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Evolutionary Hybrid Particle Swarm Optimization Algorithm for Solving NP-Hard No-Wait Flow Shop Scheduling Problems

1
Department of Computer Science and Engineering, K. L. University, Andhra Pradesh, Guntur 522502, India
2
Department of Mechanical Engineering, Walchand College of Engineering, Maharashtra, Sangli 416415, India
*
Author to whom correspondence should be addressed.
Algorithms 2017, 10(4), 121; https://doi.org/10.3390/a10040121
Submission received: 29 July 2017 / Revised: 15 October 2017 / Accepted: 19 October 2017 / Published: 28 October 2017

Abstract

:
The no-wait flow shop is a flowshop in which the scheduling of jobs is continuous and simultaneous through all machines without waiting for any consecutive machines. The scheduling of a no-wait flow shop requires finding an appropriate sequence of jobs for scheduling, which in turn reduces total processing time. The classical brute force method for finding the probabilities of scheduling for improving the utilization of resources may become trapped in local optima, and this problem can hence be observed as a typical NP-hard combinatorial optimization problem that requires finding a near optimal solution with heuristic and metaheuristic techniques. This paper proposes an effective hybrid Particle Swarm Optimization (PSO) metaheuristic algorithm for solving no-wait flow shop scheduling problems with the objective of minimizing the total flow time of jobs. This Proposed Hybrid Particle Swarm Optimization (PHPSO) algorithm presents a solution by the random key representation rule for converting the continuous position information values of particles to a discrete job permutation. The proposed algorithm initializes population efficiently with the Nawaz-Enscore-Ham (NEH) heuristic technique and uses an evolutionary search guided by the mechanism of PSO, as well as simulated annealing based on a local neighborhood search to avoid getting stuck in local optima and to provide the appropriate balance of global exploration and local exploitation. Extensive computational experiments are carried out based on Taillard’s benchmark suite. Computational results and comparisons with existing metaheuristics show that the PHPSO algorithm outperforms the existing methods in terms of quality search and robustness for the problem considered. The improvement in solution quality is confirmed by statistical tests of significance.

1. Introduction

Scheduling is an integral part of advanced manufacturing systems. Production scheduling is the arrangement of jobs to be processed on available machines under some constraints. Flow Shop, Job Shop, and Open Shop are the classical models used to solve scheduling problems. The Flow Shop Scheduling Problem (FSSP) addresses most famous machine scheduling problems of many manufacturing systems, assembly lines, and information service facilities [1,2]. Sometimes Flow Shops have no delay situations that occur in the production environment in many real-life situations where a job must be processed continuously, without any interruption, from beginning to end, in order to follow the technological order of a process, which leads to a variant with the added constraint of “no-wait” [3]. In order to maintain continuous processing of a job in No-Wait Flow Shop Scheduling (NWFSS), the start of a job by the first machine is delayed, if required, and the scheduling of such a “no-wait” constraint has attracted many researchers. A No-Wait Flow Shop Scheduling Problem (NWFSSP) has found applications in various processing industries, such as the chemical industry [4], food [5], concrete ware production [6], pharmaceuticals [7], etc. Allahverdi [8] reviewed scheduling problems with the no-wait constraint with respect to different shop environments, performance measures, setup types, and optimal scheduling criteria. Among various optimality criteria, viz. makespan, Total Flow Time (TFT), tardiness, lateness, number of tardy jobs, etc., makespan [9] and TFT [7,10] are of major interest for solving scheduling problems of no-wait type of flow shops, because makespan and TFT determine the total processing time for an entire pool of jobs and the total processing times for individual jobs respectively.
This paper addresses TFT as an objective function for solving NWFSSP. TFT is considered to be an important performance measure that, when optimized, reflects a stable or uniform utilization of resources, a rapid turn-around on jobs, and the minimization of in-process inventory [4]. The main objective of planning a production schedule is to discover the sequence of jobs, which minimizes TFT. The classical brute force method for finding such job sequences fails for large-sized problems, as computational complexity rises exponentially as n!, where “n” is number of jobs; thus, NWFSSP is treated as combinatorial optimization problem. Because of computational complexity, researchers [11,12,13] have concluded that NWFSSP with more than two machines is NP-hard. The solutions to solve such NP-hard problems consist of an approximate algorithm which uses constructive heuristics, local search methods, and metaheuristics. Generally, heuristic algorithms can obtain near-optimal solutions in an acceptable amount of time. Earlier researchers [14,15,16,17,18] have developed efficient constructive heuristic algorithms for TFT minimization; however, these attempts are not useful for identifying near optimal solutions for larger-sized problems, as these developed algorithms usually get trapped in local optima for large-problem sizes [15]. Local search methods can find the solutions, but the quality of a solution and computational time depends to a great extent on appropriate initial populations [19]. Due to the advent of computation techniques, metaheuristics can be used to solve problems in less time so that the limitation of computational complexity can be resolved through metaheuristic applications.
The field of metaheuristics, for application in combinatorial optimization problems of the scientific and industrial worlds, is growing rapidly [20]; on the other hand, attempts to use metaheuristics for solving combinatorial optimization problems began late. The recent past has witnessed a remarkable shift towards the hybridization of metaheuristics for optimization. The current trend focuses more on problem-specific approaches that lead to hybridization [21]. This paper attempts to use a metaheuristic technique, viz. Particle Swarm Optimization (PSO) algorithm, and its hybridization with Simulated Annealing (SA) to solve NWFSSP with a consideration of the Total Flow Time (TFT) of jobs as an objective criterion. Through extensive computational analysis using the well-known Taillard benchmark suite, we demonstrate that the Proposed Hybrid PSO (PHPSO) algorithm outperforms the recent best-performing algorithms available in the literature.
The remainder of this paper is organized as follows. Section 2 provides a comparative review of various metaheuristics for solving NWFSSP. Section 3 formally defines and formulates NWFSSP. Section 4 describes metaheuristic PSO and SA along with a detailed procedure for implementing the proposed metaheuristic PHPSO. Section 5 describes PHPSO on Taillard benchmark suites, and then compares the performance of the proposed metaheuristics with that of the best-so-far algorithms. Finally, concluding remarks are given in Section 6.

2. Literature Review

Various metaheuristics have been proposed for solving NWFSSP for different objective criteria. This section provides a comparative review of various metaheuristics, used by earlier researchers, for solving NWFSSP for TFT as an optimization criterion, along with various hybridization techniques for the improvement of results obtained via metaheuristics over the last decade.
Fink and Vob [3] applied different kinds of metaheuristics and constructive heuristics, such as nearest neighbor, cheapest insertion, and the pilot method, along with steepest descent (SD), Iterated Steepest Descent (ISD), Simulated Annealing (SA), and Tabu Search (TS), and examined tradeoffs between solution quality and running time. Implementation efforts showed that high-quality results were obtained in an efficient way by applying metaheuristics. Later, Gao et al. [22] proposed the Hybrid Harmony Search (HHS) algorithm using Nawaz-Enscore-Ham (NEH) [23] heuristics, and provided a solution with an appropriate balance between global exploration and local exploitation. Gao et al. [24] attempted to use the Enhanced Migrating Birds Algorithm (EMBO), based on neighborhood search heuristics, to avoid local optima. In order to improve the quality of solutions, Filho et al. [25] came up with a novel Evolutionary Clustering Search (ECS) metaheuristic approach, and found it to have better results than the method of Fink and Vob [3], Discrete Particle Swarm Optimization (DPSO) [9]. Genetic Algorithm (GA) as a metaheuristic technique was quite popular for solving optimization problems, and, later, it was observed that the solution quality was improved using a hybridization technique. Tseng and Lin [7] proposed the hybrid genetic algorithm and a novel local search scheme for improving solution qualities. Zang et al. [26] also proposed the Hybrid-Genetic Algorithm (HGA) using a new crossover operator, which helped them to argue that metaheuristics always yield better solutions than heuristics. Further, the Asynchronous Genetic Algorithm (AGA) proposed by Xu et al. [27] provided a solution for avoiding gene diversity in a short amount of time. Recently, Wang et al. [28] used a constraint-simplified mixed integer programming model and proposed the Hybridization of GA with the Neighborhood Search (H&NSGA). Although GA yields better solutions, it requires the appropriate tuning of parameters; thus, a new population-based methodology working on the principle of social behavior was introduced.
PSO is a population-based metaheuristic technique that is quite popular nowadays, because of its better solution quality achieved with less parameter tuning efforts. The comparative analysis of various metaheuristics for NWFSSP, by Bewoor et al. [29,30], advocated the effectiveness of PSO. Some remarkable metaheuristics were developed by Pan et al. [9,10] for solving NWFSSP. Pan et al. [9] developed Discrete Particle Swarm Optimization (DPSO) by considering both the makespan and total flow time minimization as optimization criteria for solving the no-wait flowshop scheduling problem. Akhshabi et al. [31] proposed a Hybrid Particle Swarm Optimization (HPSO) algorithm, based on the Memetic Algorithm (MA), and provided better solutions. On the basis of a detailed literature review, and a study of the status of research work related to use of metaheuristics for solving NWFSSP for TFT as optimization criteria, the following research gaps were identified:
  • The published literature, thus far, has primarily addressed permutation-type flow shop scheduling problems, but fewer attempts were found in the study of the “no-wait” type variant of flow shop scheduling problems (NWFSSP).
  • Earlier researchers used PSO, variants of PSO, DPSO, and the hybridization of PSO with MA for solving NWFSSP with TFT as an optimality criterion; however, the development of an efficient algorithm using the hybridization of PSO with SA and TFT as an optimality criterion for NWFSSP has not been reported, neither for small-sized jobs (n = 20, 50, and 100) nor for large-sized jobs (n = 200 and 500).
  • Investigations done by most of the researchers for solving NWFSSP have been limited to 100/200 jobs [3,9,15,16,17]. Recently, Pan-Ruiz et al. [10] tried to solve large-sized problems up to 500 jobs for the Permutation Flow Shop Scheduling Problem (FSSP) and Akhshabi et al. [31] considered NWFSSP for solving large-sized problems up to 500 jobs. Hence, the scope for further research can be clearly sensed to develop improved metaheuristics.
This provided the impetus to study the hybridization of PSO with SA in order to solve NWFSSP optimally. To know the effectiveness and efficiency of the newly-proposed algorithm, the results produced by earlier researchers, Fink and Vob [3], DPSO [9], Pan and Ruiz [10], and HPSO [31], are used for comparison with the PHPSO algorithm.

3. No-Wait Flow Shop Scheduling Problem (NWFSSP)

No-wait Flowshop scheduling has set of “n” jobs and “m” machines. The processing time p(i, j) of every job “i” on each machine “j” is given. NWFSS has additional constraint of “no-wait”, which meansthat once a job starts at the first machine, it will be processed entirely through all “m” machines without waiting in between and without any preemption. To meet this constraint, a job may be delayed at the beginning. So, in order to solve this type of problems, a delay matrix (δ) needs to be calculated [17].
Let σ = {σ1, σ2, …, σn} represent the sequence of “n” jobs to be processed on “m” machines, and δ(i, s) represent the minimum delay on the first machine between the start of job “i” and the start of job “s”. Also, let p(σi, j) represent the processing time on machine “j” of the job at the “i” position of a given sequence, and let δ(σi−1, σi) denote the minimum delay on the first machine between the start of two consecutive jobs found in the “(i − 1)” and “i” position of the sequence. Let C(σi) be the completion time of the job in the ith position of a given sequence.
For i = 1, 2, …, n and j = 1, 2, …, m
C ( σ 1 )   =   j = 1 m p ( σ 1 , j )
C ( σ 2 )   =   δ ( σ 1 , σ 2 ) + j = 1 m p ( σ 2 , j )
C ( σ i ) = k = 2 i δ ( σ k 1 , σ k ) + j = 1 m p ( σ i , j )
The formula for total flow time (TFT) is given as:
TFT = i = 1 n C ( σ 1 )
=     i = 2 n     {     k = 2 i δ ( σ k 1 , σ k ) + j = 1 m p ( σ i , j )     } + j = 1 m p ( σ 1 , j )
=     i = 2 n ( n + 1 i )   δ   ( σ i 1 , σ i ) + i = 2 n j = 1 m p ( i , j )
The delay matrix of size n × n provides all the δ(i,k) values between the start of any two consecutive jobs i and k, where i ≠ k in a given sequence of n jobs to determine the objective function value. The delay matrix δ(i,k) values are obtained from the following equation:
δ ( i , k ) = p ( i , 1 ) + max {     h = 2 r p ( i , h ) h = 1 m p ( k , h ) , 0     }               2 r m
Given the matrix of size “n” (jobs) × “m” (machines) with processing time p(i,j), it is possible to generate (n!) number of feasible sequence of solutions, denoted as F(σ), from which the optimal sequence, denoted as F(σ*), is to be chosen and can be stated as per the equation given below:
F   ( σ * )     F   ( σ )
The problem is to determine a sequence of “n” jobs which gives minimum total flow time (TFT).

4. Proposed Hybrid PSO (PHPSO) for NWFSSP

In this paper, we propose an extension of the PSO algorithm to solve NWFSSP. PHPSO essentially differs from the standard PSO in some characteristics. While designing PHPSO, suitable particles from the current population are selected and effective local search for the selected particles are carried out. Taillard benchmark suite [32] is used as the input dataset for validating results produced by PHPSO. The procedural steps for designing PHPSO are explained in detail in the subsequent sections.

4.1. Particle Swarm Optimization (PSO)

PSO is an optimization algorithm that simulates its behavior from the biological example of a flock of birds searching for food in a defined area [33]. Birds do not know where the food is, but they know at each time how far the food is, by following the nearest food strategy. PSO simulates this behavior and finds the best solution in the search space. Each particle in PSO is used to represent a single solution. The fitness value of each particle is evaluated by the objective function. The velocity of each particle provides flying direction for food. In this context, the particle reaches towards the approximate solution for the given objective function. The standard theory and procedure of PSO is well defined by Eberhard and Kennedy [34].
The algorithm is initialized with particles at random positions, and then it explores the search space to find a better solution [18]. For each iteration, each particle adjusts its velocity to follow two best solutions. The first is the cognitive part where a particle follows its own best solution found so far, called “pbest”, and the other is the current best solution of swarm, called “gbest”. On the basis of the different learning approaches of particles, PSO presents with two versions viz. the global version and the local version. In the global PSO, each particle learns from the best particle in the whole swarm, while in the local version each particle learns from the best particle in its neighborhood. Out of these two versions, the local PSO has a slower convergence speed; thus, it may adapt to a changing environment more easily which is followed exactly in the NWFSS. The new velocity is denoted by Vnew and the new position is denoted by Xnew, as stated in Equations (9) and (10):
Vnew = w*Vcurr + c1*r1*(pbest − Xcurr) + c2*r2*(gbest − Xcurr)
Xnew = Xcurr + Vnew
where w is the inertia weight, which provides balance between local and global search capabilities. The acceleration constants c1 and c2 in Equation (9) are cognitive parameters which develop the bird’s own confidence (cognitive behavior) and its confidence in the swarm (social behavior), respectively. Low values of c1 and c2 may direct particles to roam far from target regions, whereas high values may lead towards hasty movement from target regions. So, these acceleration coefficients should be appropriately adjusted. Xnew and Vnew are the new position and velocity of the particle, respectively. Xcurr and Vcurr are the current position and velocity of the particle, respectively. In the standard PSO, the new velocity of the particle is found by Equation (9), considering its previous velocity and the distance of its current position from both its own best historical position and its neighbors’ best position. Generally, the value of each component in velocity is set to the range (Vmax, −Vmax) due to which particles cannot roam excessively outside the search space. With this new velocity, the particle moves towards a new position according to Equation (10). This process stops when the user-defined terminating criterion is met.

4.2. Solution Representation

Solution representation is one of the most important issues in designing a PSO algorithm. To represent the solution, a job-permutation-based encoding scheme [2] has been used very often by earlier researchers for solving NWFSSP. However, as the position of particles in PSO is a continuous character, a standard encoding scheme of PSO cannot be adopted directly for solving NWFSSP. PSO can be effectively applied by considering dimension size as “n” for representing “n” jobs, and related particle information is represented as Xi = {x1, x2, x3, …, xn). As permutations of jobs cannot be presented with the particle alone, it is necessary to find suitable mapping between the job sequence and the position of particles in PSO. So, in this paper, the Ranked-Order-Value (ROV) rule based on the random key value [35,36] is used to determine the permutation implied by the position values xij of particle Xi. The ROV rule converts the continuous position values of particle to a discrete job permutation. This enables one to convert the continuous nature of PSO algorithms to apply to the determination of the discrete nature of problems, such as sequencing, which in turn evaluates the performance of a particle. Moreover, permutations of jobs are constructed by considering a job index, which is the rank of each position value of a particle. The ROV rule used in PHPSO handles the particle with the smallest position value first and assigns a rank value i.e., 1, and that which is observed as the smallest is assigned to that position of the particle. In the case of two or more particles with same position values, the position with the smallest dimension number is given priority and assigned a rank value first. The remaining position values are incremented by 1 and subsequently assigned the next rank values as per the dimension number. Then, the second smallest position value will be handled in the same manner. Thus, the position information of a particle is converted to corresponding job permutation σij = [j1, j2, j3, …, jn]. To demonstrate the scheme of the ROV rule, we provide a simple example in Table 1.
Let us consider that the random position values of particle (for n = 5) observed initially are Xi = {5.45, 4.22, 4.37, 5.47, 4.37}. As x1,2 = 4.22 has the smallest position value of the particle, x1,2 is prioritized first by being assigned the rank value of 1. Next, two particles x1,3 and x1,5 have equal position value i.e., 4.37. Yet, index of x1,3 is smaller as compared to x1,5. So, X1,3 is assigned the next rank value of 2 and x1,5 is incremented to rank value 3. Finally, rank values 4 and 5 are respectively assigned to x1,1 and x1,4. Thus, the job permutation σij = [4, 1, 2, 5, 3] is obtained considering the position information value of each particle and the corresponding rank assignment based on the ROV rule. In the Proposed Hybrid PSO, job permutation based local search approaches are applied rather than a direct consideration of the particle’s position information. So, it is necessary to convert the particle’s position information to a corresponding job permutation as per the ROV rule when a local search is completed. Because of the simple mechanism of the ROV rule, adjustment for a new particle position is very easy. Local search methods using the position information are handled in the same way as the process adopted for job permutation. For example, in Table 2, if the SWAP operator [2] is used as a local search operator for job permutation; the swapping of job 2 and job 4 corresponds to the swap of position values 4.22 and 4.37.

4.3. Population Initialization

The initial swarm generation is often random in the standard PSO. An initial population with a certain quality and diversity provides an efficient solution. In this paper, we propose the NEH heuristic technique [23] as an efficient population initialization procedure. In order to find NEH-based seed sequence, jobs are ordered in ascending sums of their total flow times. The partial schedules depending on the initial order are taken into account to construct a job sequence. Consider a current sequence σij = [4, 1, 2, 5, 3]; if job 4 at index “i” is the first job, then partial sequences are constructed by inserting job 4 at all indexes where “i = i + 1” of the current sequence which may appear as [1, 4, 2, 5, 3], [1, 2, 4, 5, 3], [1, 2, 5, 4, 3], and [1, 2, 5, 3, 4]. Among all these sequences, the sequence generating the minimum TFT is chosen as the current sequence for the next iteration. Thus, initial population generation with the NEH technique helps job permutation as compared to a random initial population.

4.4. Simulated Annealing(SA)

In metallurgy, the annealing process is the process where metals are cooled slowly to reach a state of low energy where they are very strong [37]. At high temperatures, the movements are random, whereas at low temperatures, little randomness is observed. Khamlichi et al. [38] used SA as a local search method for finding neighborhoods for optimizing the number of sensors and their positions in order to achieve the desired application requirements. Here, SA is used for possible job sequences leading towards minimum TFT in the context of NWFSS. SA starts a random search at a high temperature, and eventually the temperature is reduced slowly, becoming a greedy descent as it approaches to zero degrees. Random changes in the temperature not only help to escape from local minima, but also help to find low heuristic value regions. The results may be worse initially at high temperatures, but improvements can be observed gradually at lower temperatures. For the minimization of a given objective function, temperature should be reduced according the probability (P) given by the Boltzmann factor given in Equation (11):
P = e Δ E / α T
where α is the Boltzmann constant, T is the current temperature, and Δ E is the change in energy. The Boltzmann probability is a random number between 0 and 1 drawn from a uniform distribution. If the Boltzmann probability is more than the random number, the configuration is accepted. This allows the algorithm to escape from local minima. A proper initial temperature should be maintained high so that all states of the system have an equal probability of being visited.

4.5. Proposed Hybrid PSO (PHPSO) Algorithm

PHPSO algorithm is based on the solution representation by the ROV rule, population initialization with NEH-based local search and neighborhood searching through SA-based local search. The complete computational procedure of the PHPSO framework for the NWFSS PHPSO algorithm for NWFSSP (Algorithm 1) can be summarized as follows:
Algorithm 1: PHPSO for NWFSSP
Step 1: Input the total no. of jobs (n), total no. of machines (m) and processing time matrix (p).
Calculate delay matrix (δ) as per Equation (7).
Step 2:
  for i = 0 to n − 1 do
2.1
Initialize particle iwith random value (mpval) and velocity (mvelocity). Set the acceleration constants c1 and to 1.65 and 1.75 respectively; r1 and r2 both are set to the value 0.5 and inertia weight w to 0.65.
2.2
Apply ROV rule to represent random value of particle to position of particle (mpbest).
2.3
Calculate the processing sequence of job (σ) as per ROV rule.
2.4
Evaluate objective function value TFT as per Equation (6).
  end for
Step 3:
  Sort the particles with increasing order of TFT score.
Step 4: Generate initial seed sequence with NEH algorithm by following:
4.1
Consider the first sequence(σ1) of job and find TFT. Swap first position with next and compute TFT for new sequence(σ1).
4.2
for i = 1 to n do
4.2.1
Swap σi with σi+1 and find TFT
4.2.2
if TFT(σi) < TFT(σi+1) set fseq = σi
                  else fseq = σi+1
Step 5: minArr = fseq
Step 6: Calculate pbest (mpbest) of particle and gbest (pgbest) of swarm for generating the initial seed sequence.
Step 7: Select particle from the current population for local refinement;
  repeat
7.1
for i = 0 to n − 1 do
7.1.1
Update velocity and position of particle according to Equations (9) and (10), respectively.
7.1.2
Update value of particle (mpval) and apply ROV rule to find next job permutation.
7.1.3
Calculate TFT value for the updated particle.
7.1.4
If updated_TFT_value > current_TFT_value and gbest (pgbest) then
7.1.5
update pbest of particle (mpbest) and gbest (pgbest)
                  end for
            until maximum iteration count is reached.
Step 8: Select best_particle from the population for global refinement;
Step 9: Initialize initial_temperature, T as 3.0 and final_temperature, F as 0.9, and cooling rate α as 0.99.
Step 10: Initialize Best_So_Far to current state.
Step 11: while T < final_temperature do
11.1
for i = 0 to n − 1
11.1.1
Randomly perturb from the current state to a new state and calculate corresponding objective function value.
11.1.2
Update gbest depending on best particle.
11.1.3
Calculate the difference in objective function value between current and new state i.e., ΔE
11.1.4
If ΔE < 0 i.e., new state has minimum TFT, accept new state as current state. Set Best_So_Far to this new state
11.1.5
If ΔE ≥ 0, consider new state as current state with probability by invoking random number between range (0, 1).
11.1.6
Prob (accepted) = exp (−ΔE/α·T).
11.1.7
Revise T as necessary according to annealing schedule
            end for
            end while
Step 12: set gbest to Best_So_Far.
End Procedure
Thus, it can be observed that the PHPSO effectively provides a promising solution within the entire region, along with exploitation for solution improvement in sub-regions. Because of the NP-hard nature of NWFSSP, PHPSO applies local search methods which include NEH-based local search and SA-based local search. Since both exploration and exploitation are used in this algorithm, it is expected to achieve good results for NWFSSP. The results obtained through various numerical simulations and their comparisons are demonstrated in the next section.

5. Numerical Tests and Comparisons

5.1. Experimental Setup

To test the performance of the PHPSO algorithm, a computational simulation is carried out with some well-studied benchmarks. In this paper, 120 problem instances that were contributed by the Taillard dataset are selected. The Taillard benchmark dataset is composed of 12 groups containing the problems of size ranging from 20 jobs and five machines to 500 jobs and 20 machines with 10 instances of each problem size. Further, these subsets are denoted as 20 × 5 (ta001–ta010), 20 × 10 (ta011–ta020), 20 × 20 (ta021–ta030), 50 × 5 (ta031–ta040), 50 × 10 (ta041–ta050), 50 × 20 (ta051–ta060), 100 × 5 (ta061–ta070), 100 × 10 (ta071–ta080), 100 × 20 (ta081–ta090), 200 × 10 (ta091–ta100), 200 × 20 (ta101–ta110), and 500 × 20 (ta111–ta120) representing the number of jobs and machines respectively. In this paper, we used this dataset to test our PHPSO algorithm, and this test bed is treated for NWFSSP with TFT as the optimization criterion. PHPSO is coded in Java and run on Intel Core i5, 8 GB RAM, 2.20 GHz PC.

5.2. Computational and Statistical Evaluation

To compare the proposed heuristic with the existing heuristics, we carried out the experimentation by running each instance independently 10 times; for each replication we used “Average Relative Percentage Deviation” (ARPD) as a performance measure, which is popular in the scheduling literature [9,10,14,16,17]. ARPD is given by:
A R P D = 100 k i = 1 k ( H e u r i s t i C i B e s t H i ) B e s t H i
where HeuristiCi is the total flowtime obtained by any of four algorithms, and the BestHi is the lowest total flowtime obtained for that specific instance. Table 3 displays a comparative evaluation of the proposed metaheuristic, F&V [3], DPSO [9], Pan-Ruiz [10], and HPSO [31] based on ARPD for the Taillard benchmark data suite for 500 jobs.
Table 3 exhibits the results showing that the ARPD of PHPSO is significantly less compared to existing algorithms. It is also observed that, with respect to ARPD, the proposed method performs better than either of the existing methods for 103 problems out of the 120 under consideration.
To validate the significance of the proposed algorithm statistically, the results of PHPSO are compared with the results obtained by earlier developed metaheuristics viz. F&V (2003), DPSO (2008), Pan-Ruiz (2012), and HPSO (2014). To test the performance of the proposed algorithm and the best-known solutions of earlier algorithms published in the literature, a series of the paired t-test at the 95% significance level was carried out by Devore [39]. Paired t-test analyzes the differences in two observations of the mean of the results of PHPSO and the mean of existing metaheuristics. Let μD = μ1 − μ2 denote the true difference between the ARPD generated by two different algorithms. The null hypothesis is given by H0: μD = μ1 − μ2 = 0, saying that there is no difference between the ARPD generated by two algorithms when compared. The alternative hypothesis is given by H1: μD = μ1 − μ 2 ≠ 0, saying that there is a difference between the ARPD generated by two algorithms when compared. The paired t-test results on the Taillard instances are shown in Table 4, Table 5, Table 6, Table 7, Table 8 and Table 9.
The p-value is zero. Thus, the null hypothesis was rejected on behalf of the PHPSO algorithm. This indicates that the difference between TFTs generated using both algorithms are meaningful at the confidence interval (CI) of 95%. For this reason, it can be concluded that the PHPSO algorithm is superior to F&V, DPSO, Pan-Ruiz, and HPSO.
In addition to the pair-wise comparison of the metaheuristics, to observe the statistical significance of the differences between the heuristics, the means of each metaheuristic and the corresponding 95% confidence intervals are plotted in Figure 1 and Figure 2.

5.3. Comparison of Proposed Hybrid PSO(PHPSO) with Fink and Vob, DPSOVND, Pan-Ruiz, and HPSO

We report the best-known solutions termed as objective function values found so far for NWFSSP with TFT criterion for Taillard’s benchmark suite in Table 10. First, we carried out a simulation for the effectiveness of the PHPSO algorithm, and later, we compared our PHPSO with four existing metaheuristics viz. HPSO, which is an MA-based PSO by Akhshabi [31]; PRA, which is a local search-based algorithm by Pan and Ruiz [10]; the hybrid PSO based on the variable neighborhood search (DPSO) by Pan et al. [8]; and the F&V algorithm by Fink and Vob [3]. The best solution for each of the 120 Taillard dataset available in Operation Research (OR) library [32] was considered by closely examining all existing results. They are applied to these 120 benchmark instances ranging from ta001 to ta120 (i.e., for 20–500 jobs).
Table 10 shows that the proposed algorithm improves 103 out of 120 instances of the Taillard dataset. This shows that the optimal results obtained by the PHPSO are better than values obtained by various metaheuristics to date, exhibiting the effective searching quality of PHPSO.
Compared with the results by the F & V and DPSO methods, PHPSO could improve the results to a great extent, which demonstrates the noteworthy improvement by PHPSO over F & V and DPSOVND metaheuristics. Values obtained by PHPSO are better than those obtained by HPSO for almost all of the instances. So, it is concluded that our proposed NEH- and SA-based local search methods, especially their utilizations in a hybrid sense, are more effective than the variable neighborhood-based [9] and MA-based local search methods [31], especially for large-sized problems.

6. Conclusions and Future Research

In this paper, we proposed a hybridization of PSO with SA for flow shop scheduling with a no-wait constraint. The PHPSO algorithm not only applies an evolutionary search guided by the mechanism of PSO, but also it applies a local search guided by the NEH-based initial population and the mechanism of SA. Thus, both global exploration and local exploitation are balanced. The results and comparisons of the simulation demonstrate the supremacy of PHPSO in terms of searching quality and robustness of solution.
The effectiveness of the proposed method was measured by using ARPD, which is a widely used performance measure. We carried out an extensive experimental and statistical analysis and found that PHPSO has improved objective function values for 103 out of the 120 best-known solutions for Taillard’s benchmark suite. After comparing the solutions obtained through PHPSO with the solutions provided by other algorithms reported in the literature (viz. HPSO, Pan-Ruiz, DPSO, F & V algorithms), it is clearly seen that the PHPSO algorithm outperforms the existing algorithms. Hence, to the best of our knowledge, it is concluded that the PHPSO algorithm is the improved hybrid algorithm for the application of PSO to NWFSSP with a TFT criterion.

Author Contributions

Laxmi A. Bewoor has designed and developed the algorithm under the supervision and guidance of Sagar U. Sapkal and V. Chandra Prakash. They explored the applicability of this algorithm to manufacturing scheduling problem in general and no wait flow shop in particular. Accordingly, Laxmi A. Bewoor has tested the algorithm and obtained the results which are incorporated in this paper. All of them contributed for writing this paper.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Pinedo, M. Scheduling: Theory, Algorithms and Systems, 2nd ed.; Prentice-Hall: Upper Saddle River, NJ, USA, 2002. [Google Scholar]
  2. Wang, L.; Zheng, D.Z. An effective hybrid heuristic for flow shop scheduling. Int. J. Adv. Manuf. Technol. 2003, 21, 38–44. [Google Scholar]
  3. Fink, A.; Vob, S. Solving the continuous flow-shop scheduling problem by metaheuristics. Eur. J. Oper. Res. 2003, 151, 400–414. [Google Scholar] [CrossRef]
  4. Rajendran, C. A no-wait flowshop scheduling heuristic to minimize makespan. J. Oper. Res. Soc. 1994, 45, 472–478. [Google Scholar] [CrossRef]
  5. Grabowski, J.; Pempera, J. Sequencing of jobs in some production system. Eur. J. Oper. Res. 2000, 125, 535–550. [Google Scholar] [CrossRef]
  6. Raaymakers, W.; Hoogeveen, J. Scheduling multipurpose batch process industries with no-wait restrictions by simulated annealing. Eur. J. Oper. Res. 2000, 12, 6131–6151. [Google Scholar] [CrossRef]
  7. Tseng, L.; Lin, Y. A genetic local search algorithm for minimizing total flow time in the permutation flowshop scheduling problem. Int. J. Prod. Econom. 2010, 127, 121–128. [Google Scholar] [CrossRef]
  8. Allahverdi, A. A survey of scheduling problems with no-wait in process. Eur. J. Oper. Res. 2016, 255, 665–686. [Google Scholar] [CrossRef]
  9. Pan, Q.; Tasgetiren, F.; Liang, Y. A discrete particle swarm optimization algorithm for the no-wait flowshop scheduling problem. Comput. Oper. Res. 2008, 35, 2807–2839. [Google Scholar] [CrossRef]
  10. Pan, Q.; Ruiz, R. Local search methods for the flowshop scheduling problem with flowtime minimization. Eur. J. Oper. Res. 2012, 222, 31–43. [Google Scholar] [CrossRef]
  11. Rock, H. The three-machine no-wait flow shop is NP-complete. J. ACM 1984, 31, 336–345. [Google Scholar] [CrossRef]
  12. Garey, M.; Johnson, D. Computers and Intractability, a Guide to the Theory of NP-Completeness, 4th ed.; Freeman: New York, NY, USA, 1979. [Google Scholar]
  13. Graham, R. Optimization and approximation in deterministic sequencing and scheduling. Ann. Discret. Math. 1979, 5, 287–326. [Google Scholar]
  14. Rajendran, C. A heuristic for scheduling in flowshop and flowline-based manufacturing cell with multicriteria. Int. J. Prod. Res. 1994, 32, 2541–2558. [Google Scholar] [CrossRef]
  15. Bertolissi, E. Heuristic algorithm for scheduling in the no-wait flow-shop. J. Mater. Process. Technol. 2000, 107, 459–465. [Google Scholar] [CrossRef]
  16. Aldowaisan, T.; Allahverdi, A. New heuristics for m -machine no-wait flowshop to minimize total completion time. Int. J. Manag. Sci. 2004, 32, 345–352. [Google Scholar] [CrossRef]
  17. Sapkal, S.; Laha, D. A heuristic for no-wait flow shop scheduling. Int. J. Adv. Manuf. Technol. 2013, 68, 1327–1338. [Google Scholar] [CrossRef]
  18. Tasgetiren, M.; Pan, Q.; Kizilay, D.; Gao, K. A variable block insertion heuristic for the blocking flowshop scheduling problem with total flowtime criterion. Algorithms 2016, 9, 71. [Google Scholar] [CrossRef]
  19. Liu, J.; Reeves, C. Constructive and composite heuristic solutions to the P//∑Ci scheduling problem. Eur. J. Oper. Res. 2001, 132, 439–452. [Google Scholar] [CrossRef]
  20. Blum, C.; Roli, A. Metaheuristics in combinatorial optimization: Overview and conceptual comparison. Appl. Soft Comput. J. 2003, 35, 268–308. [Google Scholar] [CrossRef]
  21. Blum, C.; Puchinger, J.; Raidl, G.; Roli, A. Hybrid metaheuristics in combinatorial optimization: A survey. Appl. Soft Comput. J. 2011, 11, 4135–4151. [Google Scholar] [CrossRef]
  22. Gao, K.; Pan, Q.; Li, J.; Wang, Y.; Liang, J. A hybrid harmony search algorithm for the no-wait flow-shop scheduling problems. Asia-Pac. J. Oper. Res. 2012, 29, 12500–12512. [Google Scholar] [CrossRef]
  23. Nawaz, M.; Enscore, E., Jr.; Ham, I. A Heuristic Algorithm for the m-Machine, n-Job Flow-shop Sequencing Problem. Int. J. Manag. Sci. 1983, 11, 91–95. [Google Scholar] [CrossRef]
  24. Gao, K.; Suganthan, P.; Chua, T. An enhanced migrating birds optimization algorithm for no-wait flow shop scheduling problem. In Proceedings of the IEEE Symposium on Computational Intelligence in Scheduling, Singapore, 16–19 April 2013; pp. 9–13. [Google Scholar]
  25. Filho, G.; Nagano, M.; Lorena, L. Hybrid Evolutionary Algorithm for Flowtime Minimization in No-Wait Flowshop Scheduling; Lecture Notes in Computer Science; Springer: Berlin, Germany, 2007; Volume 4827, pp. 1099–1109. [Google Scholar]
  26. Zhang, Y.; Li, X.; Wang, Q. Hybrid genetic algorithm for permutation flowshop scheduling problems with total flowtime minimization. Eur. J. Oper. Res. 2009, 196, 869–876. [Google Scholar] [CrossRef]
  27. Xu, X.; Xu, Z.; Gu, X. An asynchronous genetic local search algorithm for the permutation flowshop scheduling problem with total flowtime minimization. Expert Syst. Appl. 2011, 38, 7970–7979. [Google Scholar] [CrossRef]
  28. Wang, F.; Rao, Y.; Tang, Q. A hybrid intelligence algorithm for no-wait flow shop scheduling. Adv. Mater. Res. 2013, 712, 2447–2451. [Google Scholar] [CrossRef]
  29. Bewoor, L.; Chandra Prakash, V.; Sapkal, S. Comparative analysis of metaheuristic approaches for m-machine no-wait flow shop scheduling for minimizing total flow time with stochastic input. Int. J. Eng. Technol. 2016, 8, 3021–3026. [Google Scholar] [CrossRef]
  30. Bewoor, L.; Chandra Prakash, V.; Sapkal, S. Comparative analysis of metaheuristic approaches for makespan minimization for no-wait flow shop scheduling problem. Int. J. Electr. Comput. Eng. 2017, 7, 31–37. [Google Scholar] [CrossRef]
  31. Akhshabi, M.; Tavakkoli-Moghaddam, R.; Rahnamay-Roodposhti, F. A hybrid particle swarm optimization algorithm for a no-wait flow shop scheduling problem with the total flow time. Int. J. Adv. Manuf. Technol. 2014, 70, 1181–1188. [Google Scholar] [CrossRef]
  32. Taillard, E. Benchmarks for basic scheduling problems. Eur. J. Oper. Res. 1993, 64, 278–285. [Google Scholar] [CrossRef]
  33. Vannucci, P. ALE-PSO: An adaptive swarm algorithm to solve design problems of laminates. Algorithms 2009, 2, 710–734. [Google Scholar] [CrossRef]
  34. Eberhard, R.; Kennedy, J. A new optimizer using particle swarm theory. In Proceedings of the Sixth International Symposium on Micro Machine and Human Science, Nagoya, Japan, 4–6 October 1995; pp. 39–43. [Google Scholar]
  35. Bean, J. Genetic algorithms and random keys for sequencing and optimization. ORSA J. Comput. 1994, 6, 154–160. [Google Scholar] [CrossRef]
  36. Liu, B.; Wang, L.; Jin, Y.-H. An effective PSO-based memetic algorithm for flowshop scheduling. IEEE Trans. Syst. Man Cybern. Part B 2007, 37, 18–27. [Google Scholar] [CrossRef]
  37. Jarboui, B.; Ibrahim, S.; Siarry, P.; Rebai, A. A combinatorial particle swarm optimization for solving permutation flowshop problems. Comput. Ind. Eng. 2008, 54, 526–538. [Google Scholar] [CrossRef]
  38. Khamlichi, Y.; Tahiri, A.; Abtoy, A.; Bulo, I.; Lozano, F. A hybrid algorithm for optimal wireless sensor network deployment with the minimum number of sensor nodes. Algorithms 2017, 10, 80. [Google Scholar] [CrossRef]
  39. Devore, J. Probability and Statistics for Engineering and the Sciences, 9th ed.; Brooks Cole: Pacific Grove, CA, USA, 2016. [Google Scholar]
Figure 1. Means and 95% confidence intervals for different algorithms for ta001 to ta110.
Figure 1. Means and 95% confidence intervals for different algorithms for ta001 to ta110.
Algorithms 10 00121 g001
Figure 2. Means and 95% confidence intervals for different algorithms for ta001 to ta120.
Figure 2. Means and 95% confidence intervals for different algorithms for ta001 to ta120.
Algorithms 10 00121 g002
Table 1. Representation of the solution of position information of each particle and the corresponding ROV for the corresponding job permutation.
Table 1. Representation of the solution of position information of each particle and the corresponding ROV for the corresponding job permutation.
Dimension12345
xij5.454.224.375.474.37
Job Permutation41253
Table 2. Job permutations and the corresponding position information after swapping job2 and job4 for a swap-based local search.
Table 2. Job permutations and the corresponding position information after swapping job2 and job4 for a swap-based local search.
Dimension12345
xij4.374.225.455.474.37
Job Permutation21453
Table 3. Comparison of performance of the existing metaheuristics and PHPSO.
Table 3. Comparison of performance of the existing metaheuristics and PHPSO.
InstancesF&VDPSOPAN-RUIZHPSOPHPSOInstancesF&VDPSOPAN-RUIZHPSOPHPSO
ta0010.66020.66020.48640.48640ta0620.34530.32430.07790.07790
ta0020.83780.83780.61420.61420ta0630.3410.32530.08020.08020
ta0030.30020.30020.09310.09310ta0640.36140.3530.1120.1120
ta0040.57110.57110.35050.35050ta0650.25680.24310.03160.03160
ta0050.66420.66420.46990.46990ta0660.25170.2355000.6682
ta0060.82260.82260.5430.5430ta0670.410.3990.12940.12940
ta0070.74120.74120.50320.50320ta0680.26380.246000.578
ta0080.20860.20860.05660.05660ta0690.27510.26060.02760.02760
ta0090.63690.63690.42810.42840ta0700.24290.2272000.5659
ta0100.27290.27290.07470.07470ta0710.59330.58180.1520.15190
ta0110.4490.4490.20210.20210ta0720.69130.67360.17620.17620
ta0121.48441.48441.11641.11640ta0730.62870.61430.15620.15620
ta0131.46341.46341.13261.13260ta0740.60340.5860.1430.1430
ta0140.49450.49450.25710.25710ta0750.73860.7250.23680.23680
ta0151.28171.28170.83730.83730ta0760.56320.54740.07720.07720
ta0161.78661.78661.43641.43640ta0770.45280.44130.030.030
ta0171.86151.86151.39511.39510ta0780.74470.73250.26250.26250
ta0180.94090.94090.62620.62620ta0790.66510.65260.20630.20630
ta0190.67160.67160.4460.4460ta0800.61510.59870.14640.14640
ta0201.18561.18560.89440.89440ta0811.28811.2720.48750.48750
ta0213.4593.4592.88442.88440ta0821.13951.12950.4130.4130
ta0223.08423.08422.43372.43370ta0831.26021.25260.4870.4870
ta0232.77162.77162.33922.33920ta0841.35221.33870.55610.55610
ta0242.42082.42081.79121.79120ta0851.04421.02610.35560.35560
ta0251.22281.22280.96890.96890ta0861.17551.16040.43520.43520
ta0262.93892.93892.32632.32630ta0871.24911.23960.46280.46280
ta0271.55791.55791.12321.12320ta0880.50350.4935000
ta0283.14383.14382.632.630Ta0891.08521.07010.38240.38240
ta0293.06633.06632.48292.48290ta0900.37830.5000.4575
ta0302.67762.67762.1282.1280ta0910.60290.57610.10250.10250
ta0310.53090.52420.3050.3050ta0920.46590.428000.7522
ta0320.39040.38160.13450.13450ta0930.44760.4266000.7405
ta0330.36330.36020.10.10ta0940.49360.46640.03340.03340
ta0340.36820.36380.12830.12830ta0950.46350.4351000.7278
ta0350.4320.4250.18370.18370ta0960.4990.4649000.7416
ta0360.40440.40140.160.160ta0970.46380.4366000.5638
ta0370.33790.33470.1250.1250ta0980.55860.5230.06210.06210
ta0380.39470.3910.13570.13570ta0990.56590.53880.0660.0660
ta0390.39350.39320.15720.15720ta1000.59110.5580.07790.07790
ta0400.45710.45010.19530.19530ta1010.98730.96360.21220.21220
ta0410.31320.3092000.1271ta1020.36670.6266000.7785
ta0421.14021.13460.57240.57240ta1030.63470.60840.012400
ta0430.6360.63460.24030.24030ta1041.04781.01690.24250.24250
ta0440.93190.9290.47090.47090ta1050.65180.6373000.7306
ta0451.19791.19510.64460.64460ta1061.09661.06660.25760.25760
ta0460.81670.81370.39650.39650ta1070.6450.6233000
ta0470.31340.3142000.1592ta1080.96110.94640.19550.18520
ta0480.99510.99530.50390.50390ta1091.08531.05520.26420.26420
ta0490.93340.93140.49520.49520ta1100.63680.6159000
ta0500.84610.84550.43180.43180ta111--0.070.070
ta0511.74981.7521.00191.00190ta112--0.050.050
ta0520.37520.37310.0180.0180ta113--0.050.050
ta0531.5921.59030.88410.88410ta114--0.050.050
ta0541.68931.69151.00191.00190ta115--0.070.070
ta0561.20871.20870.64630.64630ta116--0.060.060
ta0570.36090.3596000.0642ta117--0.020.020
ta0581.82761.8231.06041.06040ta118--0.110.10
ta0590.48150.48070.09230.09230ta119--000.99
ta0600.39340.39340.02570.02570ta120--001.09
ta0610.32360.30510.08820.08820Avg.0.90110.89550.42440.42410.0999
Table 4. Paired t-test for H0 = PHPSO = HPSO vs. H1 = PHPSO ≠ HPSO on the best-known solutions.
Table 4. Paired t-test for H0 = PHPSO = HPSO vs. H1 = PHPSO ≠ HPSO on the best-known solutions.
AlgorithmNMeanStDevSE Mean
PHPSO110282,949380,23636,254
HPSO-2014110319,512405,33738,647
Difference110−36,563.360,259.75745.5
95% CI for mean difference: (−47,950.8, −25,175.9); t-test of mean difference = 0 (vs. not = 0): t-value = −6.36, p-value = 0.000.
Table 5. Paired t-test for H0 = PHPSO = Pan-Ruiz vs. H1 = PHPSO ≠ Pan-Ruiz on the best-known solutions.
Table 5. Paired t-test for H0 = PHPSO = Pan-Ruiz vs. H1 = PHPSO ≠ Pan-Ruiz on the best-known solutions.
AlgorithmNMeanStDevSE Mean
PHPSO110282,949380,23636,254
Pan+Ruiz-2012110319,750405,88938,700
Difference110−36,801.560,548.45773.1
95% CI for mean difference: (−48,243.5, −25,359.5); t-test of mean difference = 0 (vs. not = 0): t-value = −6.37, p-value = 0.000.
Table 6. Paired t-test for H0 = PHPSO = DPSO vs. H1= PHPSO ≠ DPSO on the best-known solutions
Table 6. Paired t-test for H0 = PHPSO = DPSO vs. H1= PHPSO ≠ DPSO on the best-known solutions
AlgorithmNMeanStDevSE Mean
PHPSO110282,949380,23636,254
DPSO-2008110472,275637,37760,771
Difference110−189,326272,43825,976
95% CI for mean difference: (−240,810, −137,843); t-test of mean difference = 0 (vs. not = 0): t-value = −7.29, p-Value = 0.000.
Table 7. Paired t-test for H0 = PHPSO = F&V vs. H1 = PHPSO ≠ F&V on the best-known solutions.
Table 7. Paired t-test for H0 = PHPSO = F&V vs. H1 = PHPSO ≠ F&V on the best-known solutions.
AlgorithmNMeanStDevSE Mean
PHPSO110282,949380,23636,254
F&V-2003110478,400647,51761,738
Difference110−195,451267,28125,484
95% CI for mean difference: (−248, 238, −141, 755); t-test of mean difference = 0 (vs. not = 0): t-value = −7.26, p-Value = 0.000.
Table 8. Paired t-test for H0 = PHPSO = HPSO vs. H1 = PHPSO ≠ HPSO on the best-known solutions.
Table 8. Paired t-test for H0 = PHPSO = HPSO vs. H1 = PHPSO ≠ HPSO on the best-known solutions.
AlgorithmNMeanStDevSE Mean
PHPSO120795,6571,746,744159,455
HPSO-2014120853,4941,820,273166,167
Difference120−57,837.6106,077.39683.5
95% CI for mean difference: (−77,011.8, −38,663.3); t-test of mean difference = 0 (vs. not = 0): t-value = −5.97, p-Value = 0.000.
Table 9. Paired t-test for H0 = PHPSO = Pan-Ruiz vs. H1 = PHPSO ≠ Pan-Ruiz on the best-known solutions.
Table 9. Paired t-test for H0 = PHPSO = Pan-Ruiz vs. H1 = PHPSO ≠ Pan-Ruiz on the best-known solutions.
AlgorithmNMeanStDevSE Mean
PHPSO120795,6571,746,744159,455
Pan+Ruiz-2012120854,6961,823,535166,465
Difference120−59,039.3109,396.39986.5
95% CI for mean difference: (−78, 813.5, −39,265.1); t-test of mean difference = 0 (vs. not = 0): t-value = −5.91, p-Value = 0.000.
Table 10. New objective function values for Taillard’s benchmarks treated as NWFSS with TFT criterion.
Table 10. New objective function values for Taillard’s benchmarks treated as NWFSS with TFT criterion.
InstanceF&VDPSOPan+RuizHPSOPHPSOInstanceF&VDPSOPan+RuizHPSOPHPSO
ta00115,67415,67414,03314,03310,841ta061308,052303,750253,266253,266232,745
ta00217,25017,25015,15115,15111,386ta062302,386297,672242,281242,281224,780
ta00315,82115,82113,30113,30112,168ta063295,239291,782237,832237,832220,164
ta00417,97017,97015,44715,44711,438ta064278,811277,093227,738227,738204,798
ta00515,31715,31713,52913,52910,204ta065292,757289,554240,301240,301232,933
ta00615,50115,50113,12313,12311,505ta066290,819287,055232,342232,342232,342
ta00715,69315,69313,54813,54813,548ta067300,068297,731240,366240,366212,821
ta00815,95515,95513,94813,94811,394ta068291,859287,754230,945230,945230,945
ta00916,38516,38514,29514,29812,010ta069307,650304,131247,921247,921241,266
ta01015,32915,32912,94312,94312,943ta070301,942298,119242,933242,933242,933
ta01125,20525,20520,91120,91117,395ta071412,700409,715298,385298,358259,015
ta01226,34226,34222,44022,44020,603ta072394,562390,417274,384274,384233,285
ta01322,91022,91019,83319,83315,300ta073405,878402,274288,114288,114249,201
ta01422,24322,24318,71018,71014,883ta074422,301417,733301,044301,044263,386
ta01523,15023,15018,64118,64116,146ta075400,175397,049284,681284,681230,167
ta01622,01122,01119,24519,24517,899ta076391,359387,398269,686269,686250,354
ta01721,93921,93918,36318,36317,667ta077394,179391,057279,463279,463271,318
ta01824,15824,15820,24120,24119,447ta078402,025399,214290,908290,908230,425
ta01923,50123,50120,33020,33020,059ta079416,833413,701301,970301,970250,337
ta02024,59724,59721,32021,32021,254ta080410,372406,206291,283291,283254,082
ta02138,59738,59733,62333,62329,656ta081562,150558,199365,463365,463245,683
ta02237,57137,57131,58731,58729,199ta082563,923561,305372,449372,449263,582
ta02338,31238,31233,92033,92030,158ta083562,404560,530370,027370,027248,834
ta02438,80238,80231,66131,66131,343ta084562,918559,690372,393372,393239,313
ta02539,01239,01234,55734,55729,551ta085556,311551,388368,915368,915272,137
ta02638,56238,56232,56432,56429,790ta086562,253558,356370,908370,908258,445
ta02739,66339,66332,92232,92225,506ta087574,102571,680373,408373,408255,264
ta02837,00037,00032,41232,41228,929ta088578,119574,269384,525384,525384,525
ta02939,22839,22833,60033,60029,647ta089564,803560,710374,423374,423270,858
ta03037,93137,93132,26232,26229,314ta090522,798568,927379,296379,296379,296
ta03176,01675,68264,80264,80249,655ta0911,521,2011,495,7301,046,3141,046,314949,025
ta03283,40382,87468,05168,05159,984ta0921,516,0091,476,8631,034,1951,034,1951,034,195
ta03378,28278,10363,16263,16257,420ta0931,515,5351,493,5021,046,9021,046,9021,046,902
ta03482,73782,46768,22668,22660,470ta0941,489,4571,462,3001,030,4811,030,481997,214
ta03583,90183,49369,35169,35158,590ta0951,513,2811,483,8941,034,0271,034,0271,034,027
ta03680,92480,74966,84166,84157,620ta0961,508,3311,474,0001,006,1951,006,1951,006,195
ta03778,79178,60466,25366,25358,893ta0971,541,4191,512,8611,053,0511,053,0511,053,051
ta03879,00778,79664,33264,33256,646ta0981,533,3971,498,3301,044,8751,044,875983,816
ta03975,84275,82562,98162,98154,424ta0991,507,4221,481,2831,026,1371,026,137962,641
ta04083,82983,43068,77068,77057,533ta1001,520,8001,489,2181,030,2991,030,299955,843
ta041114,398114,05187,11487,11487,114ta1012,012,7851,988,7721,227,7331,227,733949,025
ta042112,725112,42782,82082,82082,820ta1022,057,4092,025,5611,245,2711,245,2711,245,271
ta043105,433105,34579,93179,93164,446ta1032,050,1692,017,2161,269,6731,254,1621,254,162
ta044113,540113,36786,44686,44668,770ta1042,040,9462,010,1211,238,3491,238,3491,238,349
ta045115,441115,29586,37786,37762,523ta1052,027,1382,009,2991,227,2141,227,2141,227,214
ta046112,645112,45986,58786,58762,005ta1062,046,5422,017,2401,227,6041,227,604976,118
ta047116,560116,63188,75088,75088,750ta1072,045,9062,018,9451,243,7071,243,7071,243,707
ta048115,056115,06586,72786,72767,669ta1082,044,2182,028,8611,246,1231,235,460983,816
ta049110,482110,36785,44185,44167,144ta1092,037,0402,007,6781,234,9361,234,936962,641
ta050113,462113,42787,99887,99861,460ta1102,046,9662,020,8061,250,5961,250,596955,843
ta051172,845172,981125,831125,83162,857ta111----6,698,6566,698,6566,263,859
ta052161,092160,836119,247119,247117,137ta112----6,770,7356,723,5486,413,646
ta053160,213160,104116,459116,459101,810ta113----6,739,6456,739,6456,437,528
ta054161,557161,690120,261120,261100,074ta114----6,785,9916,743,5986,432,538
ta055167,640167,336118,184118,184114,468ta115----6,729,4686,729,4686,312,830
ta056161,784161,784120,586120,586103,248ta116----6,724,0856,724,0856,361,035
ta057167,233167,064122,880122,880122,880ta117----6,691,4686,691,4686,539,854
ta058168,100167,822122,489122,489119,449ta118----6,783,9166,755,4896,126,127
ta059165,292165,207121,872121,872111,571ta119----6,711,3056,711,3056,711,305
ta060168,386168,386123,954123,954120,849ta120----6,755,7226,755,7226,755,722
Bold values indicate improvement over existing metaheuristics.

Share and Cite

MDPI and ACS Style

Bewoor, L.A.; Chandra Prakash, V.; Sapkal, S.U. Evolutionary Hybrid Particle Swarm Optimization Algorithm for Solving NP-Hard No-Wait Flow Shop Scheduling Problems. Algorithms 2017, 10, 121. https://doi.org/10.3390/a10040121

AMA Style

Bewoor LA, Chandra Prakash V, Sapkal SU. Evolutionary Hybrid Particle Swarm Optimization Algorithm for Solving NP-Hard No-Wait Flow Shop Scheduling Problems. Algorithms. 2017; 10(4):121. https://doi.org/10.3390/a10040121

Chicago/Turabian Style

Bewoor, Laxmi A., V. Chandra Prakash, and Sagar U. Sapkal. 2017. "Evolutionary Hybrid Particle Swarm Optimization Algorithm for Solving NP-Hard No-Wait Flow Shop Scheduling Problems" Algorithms 10, no. 4: 121. https://doi.org/10.3390/a10040121

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop